Tailoring a convolutional neural network towards spectral quantification and reconstruction of simulated 7T brain FID-MRSI data

The abstract investigates the possibilities of using convolutional neural networks for enhanced and robust spectral quantification of simulated 7T FID-MRSI brain spectra. The proposed network architecture predicts wavelet parameters for baseline correction, as well as spectral parameters and metabolite amplitudes for spectral reconstruction. The potential reduction in quantification times could thus mitigate some disadvantages of current MRSI processing techniques.